Potholes along with speed bumps have been a cause of worry for motorist for a long time. Recent reports show that in India there are more than 10,000 accidents due to potholes and bumps. In this paper we attempt to identify the road surface by classifying it into pothole, speed bump and normal road based on image data. The method of classifying the road surface from the images using convolution neural networks, ResNet-50 is discussed. Initially the images are manually classified into the three classes and these are used to train the neural network, we were able to achieve a true positive rate of 88.9%. In the second phase we pass the image to object detection neural network to detect the precise location of the speed bump. This was achieved using the YOLO algorithm for object detection. This work can be extended to alert the driver and tune the suspension to make the ride more comfortable based on road preview using a camera.
Pothole and Bump detection using Convolution Neural Networks
2019-12-01
2388813 byte
Conference paper
Electronic Resource
English
POTHOLE AND SPEED BUMP DETECTION BASED ON VEHICLE'S BEHAVIORS USING COMPUTER VISION
European Patent Office | 2024
|